Software isn’t dying. It’s being absorbed, the same way it’s been absorbed multiple times before.
Every few years the tech industry discovers a new paradigm and declares everything before it obsolete. Right now it’s agents. The narrative goes: autonomous AI will replace SaaS, apps will become unnecessary, and we’ll all just talk to a model that does everything. Pages like deathbyclawd.com catalogue the casualties: “Your SaaS could be replaced by a Skill.md.”
Some of it will be true. A lot of it won’t. We’ve seen this movie before.
Aww shit, here we go again
OpenClaw’s rise is the best window into what the agent era actually looks like right now. People are automating real workflows. Someone bought a car through an agent. Someone else lost all their emails. The gap between the demos and reality is enormous, both exciting and a letdown in equal measure.
This should feel familiar. In 1999, the internet was going to replace every business. Pets.com would destroy pet stores. Webvan would end grocery shopping. The infrastructure wasn’t ready, the unit economics didn’t work, and the gap between “technically possible” and “reliably useful” was vast.
The dot-com bubble popped. But the internet didn’t go away, it quietly became the foundation for everything that came after. Amazon survived. Google emerged. The boring infrastructure companies that came afterwards (AWS, Stripe, Twilio) became the real winners, not the flashy consumer apps that got the early press.
The AI bubble will likely follow the same arc. Not because the promise of AI isn’t real, it obviously is, but because the distance between “an agent can do this in a demo” and “an agent can do this reliably at scale” is the same distance that killed Webvan.
The pattern across eras
If you zoom out far enough, software has gone through roughly seven generations:
Note: These “eras” of software are simply how I see things, there’s no universally agreed timeline, so your perspective may differ.
Each era takes the previous one’s innovation and builds on top of it. Standalone software replaced pen and paper. Client-server connected those standalone machines. Cloud moved the compute off your desk entirely. Mobile put it in your pocket. ML made it adaptive. Copilots made it generative.
But here’s the thing people miss: nothing actually died.
Excel is still here. Email is still here. Databases, REST APIs, web forms, all still here. Each era didn’t replace the previous one. It absorbed it. The “app” of one generation became the “API call” of the next.
Salesforce didn’t kill the database, it wrapped it. Slack didn’t kill email, it absorbed a subset of email’s function. Google Docs didn’t kill files, it moved them to someone else’s computer. Every generation’s revolutionary product becomes the next generation’s plumbing.
Build infrastructure not wrappers.
If the pattern holds, and I am convinced it does, agents won’t kill software.
They’ll call it. The surface-level SaaS apps that are essentially a form on top of a database? Yes, those are vulnerable. If your entire product is “collect inputs, run a query, display results,” an agent with an API key can replicate that. The deathbyclawd.com crowd is right about these.
But the software that has earned trust through reliability, that handles edge cases at scale, that manages real money or real compliance, that software becomes more valuable in an agentic world, not less.
Stripe doesn’t become obsolete because of AI, it becomes the payment service that every agent uses. The lesson from every previous era shift is the same: the winners didn’t build the flashy new interface.
They built what the next layer needed to call. File formats in the standalone era. Protocols in client-server. APIs in cloud. In the agentic era, the winners will build reliable services that agents can use.
Not the agent itself, but the things the agent reaches for.
The real question isn’t “which apps will agents replace?” It’s “which apps will agents use?” The current software layer isn’t dying. It’s just becoming infrastructure. Again.
Infrastructure which enables the “Just In Time” software era
I see the next shift as just-in-time software: interfaces assembled for the task at hand, wired to data sources and actions behind the scenes.
You might never “go to Booking.com” as the first step. You’d ask your agent to plan a trip, and it would orchestrate across Booking.com, flight trackers, your calendar, and, if you want, your partner’s availability, whether that means reading their calendar or coordinating with their agent.
The booking sites won’t disappear, they’ll become the services agents call.
The shift from desktop first to mobile first left behind the companies that didn’t adapt to the new paradigm.
The Agentic first era will do the same.